Data source
Label Studio
Checking connection…
Training target
Kaggle GPU
Checking connection…
Run pipeline
Select a project, then click Train. The pipeline uses that project's Label Studio project ID and dataset slug.
Pipeline progress
No job running
No job yet. Click Train or open the Jobs tab.
Job history
Shows jobs for the selected project only. Refreshing the page keeps the logs.
| Job | Kind | Status | Started | Finished | |
|---|---|---|---|---|---|
| No jobs yet. | |||||
Trained model
YOLO weights
Checking the training kernel for best.pt…
No trained weights yet. Run Train first, then wait until the Kaggle notebook finishes. Output files (best.pt / last.pt) appear here when the kernel status is complete.
Download model
Waiting to download best.pt…
- Speed
- —
- Elapsed
- —
- Remaining
- —
Test an image
Upload a JPG or PNG, then run locally (cached best.pt) or on Kaggle. Mask projects use SegFormer; box projects use YOLO. Testing is disabled while the training kernel is running or not complete.
Predict progress
No predict job running
- Speed
- —
- Elapsed
- —
- Remaining
- —
Result
Run a local or Kaggle predict to see boxes here.
No prediction yet.
| Class | Conf | xyxy |
|---|
HTTP
API
Callers outside this console can run the same image test. Jobs use the existing local cache + YOLO or Kaggle predict flow. Only one job runs at a time.
Auth
Start a local test
Poll the job
Response shapes
Notes
Upload an image and click Run locally.
Catalog
Training projects
Each project is tied to a Label Studio project ID and a Kaggle dataset slug. Connection details live in Settings. DeepCrack and MyCracks Refined use export type Masks (brush) and train SegFormer — not YOLO boxes.